EARTHQUAKE CLUSTERING DUE

TO STRESS INTERACTIONS
RMS Special Report

EXECUTIVE SUMMARY
In the development of probabilistic catastrophe models for earthquake risk, Risk Management Solutions (RMS)
uses historical earthquake records and the best available data from national geological institutions to determine
the expected recurrence of various magnitude earthquakes on a seismic source. For example, when the past
earthquake history is known for a given fault, time-dependent rates that take into account a faults seismic
cycle are utilized to give an indication of the degree of tectonic loading on that fault, and hence the likelihood
of fault rupture. RMS currently incorporates time-dependent probabilities in areas such as the California region
of the U.S., Japan, Mexico, Turkey, New Zealand, Israel, and Chile.
However, to consider earthquake clusteringthe occurrence of earthquakes close together in time and space
interaction-based probabilities must be calculated from the theory of stress transfer and the phenomenon of
stress interactions, where stress release on one fault can increase or decrease stress on a nearby fault,
potentially triggering or retarding an earthquake, respectively. As a result, the secular (long-term) rate of
tectonic stress loading on an active fault must be known. This information is provided in the scientific literature
only for very well-studied regions, such as the U.S., Turkey, or Japan. In this report, stress transfer theory is
discussed, as are some well-known examples of earthquake clustering due to stress interactionsin the U.S.,
Turkey, and Indonesia. RMS is committed to the incorporation of the latest scientific knowledge into its models
for the assessment of earthquake risk; future U.S. and Japan earthquake model releases will consider the
impacts of potential earthquake clustering in areas of well-defined faulting.

2008 Risk Management Solutions, Inc.

INTRODUCTION
New generations of probabilistic catastrophe models need to represent possible spatio -temporal variability of
natural hazard perils. For instance, the serial clustering of hurricanes and windstorms in space and time can be
responsible for large cumulative insured losses as seen, for example, during the 2004 and 2005 Atlantic
hurricane seasons and the 1999 European windstorm season.
Also of particular interest is the understanding of large earthquakes that cluster both spatially and temporally,
with occurrences within years or tens of years rather than typical individual fault recurrence rates of hundreds
of years or longer. These changes in rates can be quantified by using an interaction-based methodology instead
of the standard Poisson process or time-dependent renewal model for earthquake recurrence. By refining
catastrophe models to reflect the possible role of earthquake interactions, a better ev aluation of the earthquake
risk can be proposed, helping insurance companies to better manage their portfolios.
The purpose of this report is to present the science of earthquake interactions and to illustrate how earthquake
clustering caused by such interactions influences the seismic hazard in regions of high exposure. It should be
noted that there exist other models of earthquake clustering, such as the double -branching, time-dependent
model proposed by Marzocchi and Lombardi (2008) and the Short Term E arthquake Probability (STEP) model
(Gerstenberger et al., 2004), among others. 1 The examples shown in this report are some of the best-known
cases of earthquake clustering due to stress interactions and each of them has been chosen for its unique
location and particular scientific and risk management interests. These examples include:

The earthquake sequence propagating along the North Anatolian Fault in Turkey, leading to the two
devastating magnitude (M) >7 earthquakes that occurred within a three-month period in 1999

The two M>8.5 earthquakes that struck Sumatra, Indonesia within a three-month period in late 2004 into
early 2005, and a subsequent earthquake sequence that continues to the present day

The 18111812 New Madrid earthquake sequence in the Central United States, in which four M>7
earthquakes occurred within a two-month period

For more information, see Proceedings of the 2008 RMS Science Symposium: Advances in Earthquake Forecasting , at
http://www.rms.com/publications.

2008 Risk Management Solutions, Inc.

EARTHQUAKE STRESS TRANSFER

It is currently well recognized that earthquakes influence the timing and location of subsequent events and that
this phenomenon can be quantified by the theory of stress transfer (e.g., Stein, 2003; King et al., 1994; King
and Cocco, 2000).

The Seismic Cycle

The state of stress on a fault is a key factor used to determine if rupture is imminent. A common view is that
once an earthquake has occurred and released stress on a fault, the fault will remain quiet until stresses in the
Earths crust have time to rebuild, typically over hundreds to thousands of years. Probabilistic earthquake
catastrophe models follow this concept, which is known as the seismic cycle (Figure 1). The mean return period
depends on the long-term rate of tectonic stress loading
t / t, which is usually assumed constant; the
failure stress f ; and the earthquake stress drop
.
Variations
in
the stress drop lead to some deviations from
0
the mean return period, a concept known as aperiodicity. In practice, event probabilities are directly
determined from the recurrence of past earthquakes by using statistical approaches, such as the Poisson
process method or a time-dependent renewal model. In the Poissonian approach, the odds of an earthquake do
not change with time, whereas in the renewal model approach, the odds increase with time based on the time
since the last event.
Stress on fault

Figure 1: Illustration of the seismic cycle, in which the stress on a fault increases due to tectonic
loading (t/t). Once the stress has reached the failure stress (f), an earthquake occurs,
releasing stress on the fault (0), and a new seismic cycle starts.

Interaction-Based Probability Calculations

An important discovery made in the mid-1990s (Reasenberg and Simpson, 1992; King et al., 1994) was that the
stress released on a fault during an earthquake does not simply dissipate; instead , it moves down the fault and
concentrates in sites nearby, typically at the tips of the rupture. On nearby f aults, at distances of kilometers to
hundreds of kilometers, the effect is an increase in stress. This stress jump will displace the one due to tectonic
stress loading assumed in the seismic cycle. Although the stress increase may be small (typically less than or
equal to about 1% of the stress released in the initial earthquake), its effects can be significant. Note that a
decrease of stress would have the opposite effect, possibly delaying the next earthquake, a phenomenon
known as quiescence. These increases and decreases in stress along a fault are described by the stress transfer
theory. As illustrated in Figure 2, positive stress transfer (i.e., an increase in stress) can temporally modify a
faults seismic cycle and advance the next earthquake by a time period T.

2008 Risk Management Solutions, Inc.

Stress on fault

Figure 2: Illustration of the theory of stress transfer. If an event strikes on a nearby fault, it may
increase the state of stress on the fault (c>0 in this example). The next event is then advanced
by a time shift (T, in years), which modifies the return period () and consequently increases
the risk on the fault.
The time shift ( T, in years) corresponding to this advance, shown in Figure 2, is determined by dividing the
local, induced stress change on the fault due to a nearby earthquake by the secular, or long-term, rate of
tectonic loading on that fault:
T= stress change on fault/long-term tectonic loading
When the state of stress of a fault is perturbed, its characteristic rate of occurrence , defined as the inverse of
the mean return period , is temporarily modified until a new earthquake occurs on the fault. The new
expected rate is determined from the mean return period minus the time shift T due to stress transfer:
=1/(- T)
with the temporary return period shorter than the mean return period when the stress is increased on the fault
( T >0). In Figure 2, this is shown as 3.
An interaction-based probability is obtained by changing the characteristic rate to or alternatively changing
the mean return period from to in the chosen probability density function (Figure 3, solid lines).
Note that the time shift T is due to a permanent change of stress on the fault. It has been proposed that
transient stress changeswhich lead to the burst of aftershocks in the days and weeks after the mainshock
can also be implemented in interaction-based probability estimates (Figure 3, dashed lines). Different methods
have been proposed to compute these transient effects (e.g., Parsons, 2005).

2008 Risk Management Solutions, Inc.

Figure 3: 1-year earthquake probability versus time, before and after the occurrence of an earthquake on
a nearby fault at T=75 years. With a characteristic return period =100 years and a time shift T =10
years, the probabilities based on a time-independent (blue) and a time-dependent (red) model are
illustrated. The time-independent model is based on the Poisson process, whereas the time-dependent
renewal model is based on a log-normal probability density function.

2008 Risk Management Solutions, Inc.

EARTHQUAKE INTERACTIONS IN TURKEY

The North Anatolian Fault in Turkey is a major active strike-slip fault along the boundary between the Eurasian
Plate and the Anatolian Plate. It is among the worlds most heavily populated fault zones (see Figure 7) and an
ideal case study for earthquake clustering and propagating rupture sequences. Earthquakes have been shown
to jump from one fault segment to another from the east end to the west end of the fault, starting in 1939 and
ending in a series of two events in 1999 (Figure 4). As shown in Figure 4, an additional earthquake occurred on
the Marmara Fault system in 1912, but this event was not part of the sequence.

Figure 4: The propagating earthquake sequence from 1939 to 1999 along the North Anatolian Fault in
Turkey, an example of earthquake clustering due to stress interactions (Source: ETOPO2 Global Digital
Elevation Model from NOAA National Geophysical Data Center)

The 1999 Clustered Events

The two most recent and destructive earthquakes in Turkey occurred on the North Anatolian Fault in 1999. The
magnitude (M) 7.6 Izmit Earthquake, also known as the Kocaeli Earthquake, occurred on August 17, 1999. Only
three months later, the M7.2 Dzce Earthquake struck 110 km (68 mi) to the east on November 12, 1999. The
series of earthquakes caused close to 19,000 fatalities with over 48,000 hospitalized injuries (Erdik, 2000). At
the time, the population in the area affected by the two earthquakes was about 20 million, one-third of the
population of the entire country. Furthermore, almost one-half of the Turkish economic infrastructure is located
in this region (Sahin and Tari, 2000). Insurance loss estimates following the Kocaeli Earthquake were
approximately US$2 billion (in 1999 dollars) (Swiss Re, 2000). The Kocaeli Earthquake affected residential parts
and city centers of many towns, in particular Izmit (Figure 5, left) but also Dzce, which was then devastated
by the second event a few months later (Figure 5, right).

2008 Risk Management Solutions, Inc.

Figure 5: Example of damage in Izmit (left) and in Dzce (right), Turkey, as a result of the 1999
earthquakes on the North Anatolian Fault (Source: U.S. Geological Survey)
Using the example of the 1999 Kocaeli and Dzce earthquake cluster, Figure 6 illustrates the principle of stress
interactions. The net stress changes due to the 1999 Kocaeli Earthquake are represented in red for a stress
increase and in purple for a stress decrease. The Dzce Earthquake occurred three months later in a region of
increased stress, as predicted by the theory of stress transfer. For the whole Turkey and Aegean region, it has
been shown that the majority of large earthquakes since the start of the 19th century (>30 events) occurred in
regions of increased stress and none have occurred in regions of clear decreased stress (Nalbant et al., 2002;
Nalbant et al., 1998; Stein et al., 1997; Parsons et al., 2000; Hubert -Ferrari et al., 2000).

Figure 6: Net stress changes due to the M7.6 August 1999 Kocaeli Earthquake, which ruptured about 140
km (87 mi) of the North Anatolian Fault (white line). Stress along the fault and in the surrounding region
decreased (purple), except at the tips of the rupture where the stress increased (red). This stress
increase led to the M7.2 Dzce Earthquake (to the east) in November 1999, which ruptured 45 km (28
mi) of the fault (black line). Westward, in the Sea of Marmara region, no earthquake has occurred to
date, although the risk has increased because of the Kocaeli Earthquake.

2008 Risk Management Solutions, Inc.

Present-Day Seismic Hazard in the Istanbul Region

Following the 1999 earthquake events, the net positive stress changes migrated to the ends of the fault (Figure
6). Arguably, the increase in stress along the Marmara Fault system, which is just south of the city of Istanbul,
has increased the risk for the city (Parsons et al., 2000; Hubert-Ferrari et al., 2000), an urban center of 10
million inhabitants (Figure 7).

Figure 7: Map illustrating the population density along the North Anatolian Fault (Source: Population
density from Landscan 2005 available at http://www.ornl.gov/landscan/ with fault network from the
Institut de Physique du Globe de Paris)
Figure 8 illustrates the present-day stress conditions in this region, computed using the most recent data
available (Armijo et al., 2005). An accurate knowledge of the historical earthquake record in the region,
including well-resolved ruptures for the 1894, 1912, and 1999 events, in combination with the secular loading
stress, gives a realistic view of the present-day seismic hazard near Istanbul.
The quantification of the net stress changes due to a recent earthquake is crucial to determine the risk linked
to clustering. However, a new earthquake will only be triggered if the fault is already sufficiently loaded (i.e.,
late in its seismic cycle). The current state of loading on the fault can only be assessed with the knowledge of
the fault history and of the secular rate of the tectonic stress loading. The North Anatolian Fault (plus the
Marmara Fault system in the Sea of Marmara) is therefore one of the most appropriate on Earth to determine
earthquake interactions. Reliable information covers almost two seismic cycles (i.e., over the past 500 years or
so), providing a unique opportunity to quantify the actual state of stress of the region.
Records kept during the Ottoman Empire provide accurate descriptions of earthquake damage that date back to
the start of the 16 th century. Today, more details on fault history are available through new field studies (e.g.,
discovery of submarine fault scarps in the Sea of Marmara using a remote operated vehicle) and proposed
models of secular stress loading based on geodetic data sets (e.g. , from GPS data).

2008 Risk Management Solutions, Inc.

Figure 8: Present-day stress conditions in the Istanbul and Sea of Marmara region. The 1894, 1912, and
1999 earthquakes have released the stress due to secular, or long-term, tectonic loading on a
consequent part of the North Anatolian Fault and Marmara Fault system. A slip deficit or seismic gap
remains close to the city of Istanbul (Note: stress changes due to tectonic loading and stress interactions
re-computed from Armijo et al., 2005).
A team from the U.S. Geological Survey quantified the seismic hazard in Istanbul using a model that
incorporated stress interactions (Parsons et al., 2000). Based on the Poisson approach, they found a 15 to 25%
chance of an earthquake damaging the city from 2000 through 2030. However, because the major faults of the
Istanbul region are likely late in their seismic cycles, the probability increases to nearly 5015% based on the
renewal model of earthquake recurrence. Furthermore, with the increase of stress due to the 1999 Kocaeli
Earthquake, the interaction-based probability is over 6015%.

2008 Risk Management Solutions, Inc.

EARTHQUAKE INTERACTIONS IN INDONESIA

The M9.3 Indian Ocean Earthquake that was initiated off the coast of Sumatra and the Andaman Islands on
December 26, 2004 is best known for its devastating tsunami, which propagated across the entire Indian Ocean
basin (see RMS, 2006 for a summary of this event). According to the U.S. Geological Survey, if both the direct
and indirect effects of the earthquake are taken into account, this event was the fourth deadliest earthquake in
recorded history, behind the 1556 Huaxian Earthquake in China, the 1976 Tangshan Earthquake in China2, and
1138 Aleppo Earthquake in Syria (USGS, 2008). Only three months later on March 28, 2005, a second major
event, the M8.7 Nias Earthquake, struck just south of the 2004 rupture, causing more than 1,000 casualties. A
tsunami was again generated, this time only causing minor damage in the far-field. Since then, several other
large (M>7) and deadly earthquakes have occurred along the Sumatra subduction zone , also known as the
Sunda Trench.
Smaller events (M<7) occurring on the Sumatran Fault, a large onshore transform (strike-slip) fault running
north to south on Sumatra Island, can also have disastrous consequences due to the proximity of this fault to
major population centers. Figure 9 illustrates the tectonics of the region, as well as the population density
throughout Sumatra and the northern end of Java.

Figure 9: Map of the Sunda Trench and Sumatran Fault off the coast and onshore, respectively, of the
Indonesian island of Sumatra (left) (Source: ETOPO2 Global Digital Elevation Model from NOAA National
Geophysical Data Center) and the population density of the Indonesian islands and Sumatra and northern
Java (right) (Source: Landscan 2005 available at http://www.ornl.gov/landscan/)

The 2004 and 2005 Clustered Events

Insured property losses from the 2004 Indian Ocean Earthquake and Tsunami were approximately US$5 billion
(in 2004 dollars) (Swiss Re, 2004). The earthquake-induced tsunami particularly affected the Indonesian
province of Aceh. Figure 10 shows satellite imagery of the city of Lhoknga before and after the event,

For more information, see The 1976 Great Tangshan Earthquake: 30-Year Retrospective , at http://www.rms.com/publications.

2008 Risk Management Solutions, Inc.

10

illustrating the tsunamis effects. The area, which sustained run-up heights of 25 m (82 ft) (EERI, 2005), was
completely destroyed.

Figure 10: Lhoknga, Aceh Province, before (left) and after (right) the 2004 Indian Ocean Tsunami
(Source: Ikonos satellite imagery on January 10, 2003 and December 29, 2004)
The second clustered event occurred three months afterward in March 2005, about 200 km (124 mi) south of
the first event, causing widespread power outages and more damage in regions already devastated by the first
earthquake and tsunami. In particular, Nias Island, which was partially destroyed by the 2004 earthquake and
tsunami (Figure 11), was in close proximity to the epicenter of the March 2005 event. As a result, when the
2005 earthquake struck, it did even more damage in the southwest part of this island than the 2004 event.

Figure 11: Damage at Sirombu, Nias Island, after the December 2004 earthquake and tsunami (Source:
California Institute of Technology)

2008 Risk Management Solutions, Inc.

11

Changes in Seismic Hazard since 2004

Table 1 lists some of the deadly earthquakes along the Sunda Trench and Sumatran Fault between 2004 and
2008.3 Events on the Sunda Trench, or Sumatra subduction zone, are generated by thrust faulting and able to
produce tsunamis, while events on the Sumatran Fault are generated onshore by strike-slip faulting. Although
events on the Sumatran Fault are 1 to 2 magnitude units smaller than subduction earthquakes, they are also
able to produce significant damage, such as the March 6, 2007 event , which killed 70 people.
Table 1. Deadly earthquakes in the Sumatra region between 2004 and 2008 (Source: USGS

http://earthquake.usgs.gov/regional/world/historical_country_mag.php#indonesia )
Magnitude

Fault

Potential Stress
Triggered Event?

Fatalities

December 26, 2004

9.3

Sunda Trench

No

283,105*

March 28, 2005

8.7

Sunda Trench

Yes

1,315

July 17, 2006

7.7

Sunda Trench

No

730

March 6, 2007

6.4

Sumatran Fault

Yes

70

8.4; 7.9

Sunda Trench

No

25

7.4

Sunda Trench

Yes

Date

September 12, 2007

February 20, 2008

* Due to earthquake ground shaking and tsunami

The net stress increase on the Sumatra subduction zone and the Sumatran Fault resulting from the four largest
earthquakes (M>7) that occurred in the region between 2004 and 2007 (see Table 1) is represented on Figure
12. The net stress increase is shown at three different time s: December 2004, July 2006, and February 2008.
The events are chronicled below.

December 2004February 2005

The M9.3 Indian Ocean Earthquake occurs, relaxing stress along a consequent part of the Sumatra subduction
zone and increasing stress southward (McCloskey et al., 2005; Pollitz et al., 2006) and on the north part of the
Sumatran Fault (McCloskey et al., 2005). The rupture extent of the future March 28, 2005 M8.7 event is
outlined in black just south of the 2004 rupture, and its northern end is located in a region of increased stress.
Field observations made before 2006 along the Sunda Trench (off the west coast of central Sumatra), show
that this portion of the subduction zone is near the end of its seismic cycle (Natawidjaja et al., 2006). It
suggests that this part of the subduction zone (shown in red in December 2004 on Figure 12) is ready to
rupture because of the long-term tectonic loading.

March 2005July 2006

The March 2005 Nias Earthquake (M8.7) occurs only three months after the 2004 event in a region of increased
stress, with its epicenter approximately 189 km (117 mi) south of the 2004 event. Stresses imposed by this
second rupture have brought the thrust segment immediately to the south closer to failure and have expanded
the area of increased stress on the Sumatran Fault (Nalbant et al., 2005). Another event (M7.7) occurs south of
Sumatra along the subduction zone in July 2006. However, there was no evidence of stress transfer and the
degree of stress loading on the fault segment was unknown.

For a complete list of Indonesian earthquakes, see

http://earthquake.usgs.gov/regional/world/historical_country_mag.php#indonesia.

2008 Risk Management Solutions, Inc.

12

Figure 12: Net stress increase on the Sumatra subduction zone and Sumatran Fault from the end of 2004
to the beginning of 2008. Yellow stars represent earthquake epicenters. Blue regions correspond to
earthquake ruptures where stress has been released. Orange to red regions correspond to stress
increases, computed for dipping thrust planes on the subduction zone and for vertical strike-slip planes
on the Sumatran Fault. In addition, areas of increased stress as seen through field observations
indicating that a fault is approaching the end of its seismic cycle are shown in dark red on the December
2004 and July 2006 graphics. Note that only stress increases are represented (see McCloskey et al., 2005
and Nalbant et al., 2005).

2008 Risk Management Solutions, Inc.

13

August 2006February 2008

A M6.4 earthquake occurs in March 2007 on the Sumatran Fault in a region where the stress has been
increased by the M8.7 2005 Nias event. Two subsequent earthquakes (M8.4 and M7.9) occur in September
2007 in the middle of the seismic gap of the Sumatra subduction zone (i.e., locked segment that has not
experienced seismic activity for a long time). There were observations that the tectonic stress loading at the
location of these two large earthquakes matched the stress released during the events of 1797 and 1833
(Natawidjaja et al., 2006; Nalbant et al., 2005). Then, in February 2008, a M7.4 earthquake struck at the
boundary between the 2004 and 2005 rupture zones on the subduction zone. It is unclear whether this
earthquake occurred in a region of increased stress or in a region relaxed since March 2005, as assumed in the
map in Figure 12, where the continuous blue coloring indicates fault rupture across entire length of the
subduction zone.
The fact that no seismic gap seemed to remain between the events of December 2004 and March 2005 (Figure
13, left) would suggest that all the stress was released in this region. However, fault heterogeneities, such as
bends, can lead to local stress perturbations (Figure 13, right). In such a case, the February 2008 event might
have been a result of stress concentration, although this process is highly difficult to quantify.

Figure 13: Rupture slip models for the December 2004 and March 2005 earthquakes (left) (Source:
USGS) and possible stress concentration at the bend between the 2004 and 2005 ruptures (right)
At present, two distinct segments of the Sumatra subduction zone are highly stressed due to stress transfer and
could generate M>7.5 earthquakes. The northern and southern parts of the Sumatran Fault are also stressed
and could potentially produce clustered events in the range of M67.

Present-Day Seismic Hazard

An important question that arises is the extent to which these potential clustered eve nts are predictable. While
accurate interaction-based probabilities can be proposed for Turkey, as described in the previous section, it is
extremely difficult to do so in the Sumatra region, where fault data is sparse. Although the effect of a jump in
stress can be determined to localize regions most likely to host a clustered event (see Figure 12), the lack of

2008 Risk Management Solutions, Inc.

14

information on past earthquake history and on the secular rate of the tectonic stress loading lead s to large
uncertainties in the computation of time shifts. However, a compilation of results from new field observations
(e.g., Pollitz et al., 2006) and other future geodetic studies should help decrease uncertainties associated with
interaction-based probability calculations in the Indonesian region.

2008 Risk Management Solutions, Inc.

15

EARTHQUAKE INTERACTIONS IN THE NEW MADRID SEISMIC ZONE

Another well-known example of earthquake clustering is the case of the New Madrid earthquake sequence that
occurred in the Central U.S. in late 1811 and early 1812. Over the span of two months, four M>7 earthquakes
occurred along the faults bordering the states of Arkansas, Tennessee, Kentucky, Missouri, and Illinois. The
first event, which occurred on December 16, 1811 with an estimated magnitude of 7.28.1, and the last event,
which occurred on February 7, 1812 with an estimated magnitude of 7.58.0, are believed to have ruptured two
fault zones defined by modern seismicity: the northeast-striking Cottonwood Grove Fault and the west-dipping
Reelfoot thrust fault, respectively (Figure 14). The second earthquake was the smallest of the sequence,
occurring the same day as the first one on December 16, 1811 with a magnitude of 7.0this event may have
been an aftershock, or may have occurred on a separate fault. The third earthquake of M7.07.8 struck on
January 23, 1812 and has been interpreted in two ways: as a strike-slip rupture on the Northeast Arm of the
New Madrid Seismic Zone, or a remotely triggered event that occurred outside of the New Madrid Seismic Zone
(Mueller et al., 2004).
Although the 18111812 earthquake sequence changed the course of the Mississippi River and destroyed entire
forests, fatalities and damage to infrastructures were low because the area was sparsely settled in the early
19 th century (Figure 15). The present-day population in the New Madrid region, however (Figure 14), includes
the cities of Memphis, Jackson, Jonesboro, and Cape Girardeau. If this earthquake sequence occurred today, it
would have devastating consequences. An analysis conducted by RMS indicates that a repeat of the February 7,
1812 event alone would cause over US$115 billion in insured losses to the surrounding region (Hall et al.,
2006).

Figure 14: The New Madrid Seismic Zone with faults highlighted by microseismicity from 1974 to 2008
(left) (Source: Center for Earthquake Research and Information (www.ceri.memphis.edu) with SRTM30
elevation model); Population distribution in the New Madrid region (right) (Source: Landscan 2005
available at http://www.ornl.gov/landscan/)

2008 Risk Management Solutions, Inc.

16

Figure 15: Woodcut depicting the destruction following the 18111812 New Madrid earthquakes (left)
(Source: State Historical Society of Missouri); and surface deformation east of Reelfoot Lake in
Tennessee, resulting from the New Madrid earthquakes (right) (Source: USGS)

Scenarios of Clustered Events

The actual impact of the 18111812 earthquake sequence is still poorly understood. In particular, the January
23, 1812 event had few documented eyewitness accounts. Furthermore, since this time, there has not been a
significant event on these faults. Due to the lack of historical information, one way to assess the current
seismic hazard in the New Madrid Seismic Zone based on interaction-based methods is to test different stress
transfer scenarios, as illustrated in Figure 16. Since return periods in such an intra-plate environment are long
approximately 550 to 1,100 years (Wesnousky et al., 1992)a Poissonian approach to earthquake recurrence
clearly cannot explain a succession of four M>7 earthquakes striking in a two-month period as observed in 1811
and 1812. It should be noted, however, that paleoliquefaction data in the New Madrid region suggests a return
period of only 500 years in this area.
By taking into account all known faults in the New Madrid Seismic Zone, interaction-based models show that a
series of two events and a series of three events are possible in many ways. For example, based on the theory
of stress transfer, there is a clear interaction between the Cottonwood Grove Fault and the Reelfoot thrust fault
that most likely led to the 18111812 cluster sequence.

2008 Risk Management Solutions, Inc.

17

Figure 16: Scenarios of clustered events in the New Madrid Seismic Zone inferred from stress changes.
The two upper rows show the net stress changes due to a variety of fault ruptures represented in white.
The two lower rows show the possible earthquake clustersa series of two or three events in sequence
(in red with arrows indicating stress transfer to other faults).

2008 Risk Management Solutions, Inc.

18

CONCLUSIONS
In certain active tectonic regions of the world, it is possible to illustrate how interaction -based earthquake
recurrence models fit nicely to the observed historical record. For example, the 1999 M>7 events in Izmit and
Dzce, Turkey, and the 20042005 M>8 events in the Indian Ocean are good examples of the triggering of
earthquakes due to stress interactions. Furthermore, the theory of stress transfer can be used in probabilistic
earthquake catastrophe models to quantify the risk associated with spatio -temporal clustering. Different
scenarios of clustered events can be explored, such as in the New Madrid Seismic Zone in the Central U.S.,
which lead to new alternatives for earthquake risk assessment.
The science of earthquake interactions is an active, continually evolving field of research. For example, another
area of ongoing work not discussed in this report is that of dynamic triggering, in which the propagating wave
front generated during an earthquake may provide the catalyst for rupture on another fault. RMS is committed
to the continual research on earthquake clustering and its impacts on recurrence and risk. Future releases o f
the RMS Simulation Platform will include the clustering of earthquake events in the U.S. , Japan, and other
applicable regions around the world.

2008 Risk Management Solutions, Inc.

19

REFERENCES
Armijo, R. et al., 2005. Submarine fault scarps in the Sea of Marmara pull -apart (North Anatolian fault):
Implications for seismic hazard in Istanbul. Geochemistry Geophysics Geosystems 6: Q06009.
Earthquake Engineering Research Institute (EERI), 2005. The Great Sumatra Earthquake and Indian Ocean
Tsunami of December 26, 2004. EERI Special Earthquake Report, March 2005.
Erdik, M., 2000. Report on 1999 Kocaeli and Dzce (Turkey) Earthquakes. Department of Earthquake
Engineering, Bogazici University, Istanbul, Turkey.
http://www.koeri.boun.edu.tr/depremmuh/eqspecials/kocaeli/Kocaelireport.pdf.
Gerstenberger, M. et al., 2004. Real-time Forecasts of Tomorrows Earthquakes in California: A New Mapping
Tool. USGS Open-File Report 2004-1390.
Hall, L. et al., 2006. Insured losses for repeats of the 1906 San Francisco and 1811/1812 New Madrid
earthquakes: how does the hazard relate to risk? Proceedings of the 100th Anniversary Earthquake
Conference: Managing Risk in Earthquake Country . SSA-000479, Disc 2, San Francisco, California.
Hubert-Ferrari, A. et al., 2000. Seismic hazard in the Marmara Sea region following the 17 August 1999 Izmit
earthquake. Nature 404 269-273.
King, G. C. P., Stein, R. S. and Lin, J., 1994. Static stress changes and the triggering of earthquakes. Bulletin
of the Seismological Society of America 84: 935-953.
King, G. C. P. and Cocco, M., 2000. Fault interaction by elastic stress changes: new clues from earthquake
sequences. Advances in Geophysics 44: 1-38.
Marzocchi, W. and Lombardi, A.M., 2008. A Double Branching Model for Earthquake Forecasting. Submitted to
Journal of Geophysical Research .
McCloskey, J., Nalbant, S. S. and Steacy, S., 2005. Earthquake risk from co -seismic stress. Nature 434: 291.
Mueller, K., Hough, S. E., and Bilham, R., 2004. Analyzing the 1811-1812 New Madrid earthquakes with recent
instrumentally recorded aftershocks. Nature 429: 284-288.
Nalbant, S. S., McCloskey, J., Steacy, S. and Barka, A. A., 2002. Stress accumulation and increased seismic risk
in eastern Turkey. Earth and Planetary Science Letters 195: 291-298.
Nalbant, S. S., Hubert, A., and King, G. C. P., 1998. Stress coupling between earthquakes in northwest Turkey
and the north Aegean Sea. Journal of Geophysical Research 103: 24469-24486.
Nalbant, S. S. et al., 2005. Earthquake risk on the Sunda trench. Nature 435: 756-757.
Natawidjaja, D.H. et al., 2006. Source parameters of the great Sumatran megathrust earthquakes of 1797 and
1833 inferred from coral microatoll. Journal of Geophysical Research 111.
Parsons, T. et al., 2000. Heightened odds of large earthquakes near Istanbul: An interactio n-based probability
calculation. Science 288: 661-665.
Parsons, T., 2005. Significance of stress transfer in time-dependent earthquake probability calculations.
Journal of Geophysical Research 110: B05S02.1-B05S02.20.
Pollitz, F. F. et al., 2006. Stress changes along the Sunda trench following the 26 December 2004 Sumatra Andaman and 28 March 2005 Nias earthquakes. Geophysical Research Letters 33: L06309.

2008 Risk Management Solutions, Inc.

20

Reasenberg, P.A., and Simpson, R.W., 1992. Response of Regional Seismicity to the Static Stress Change
Produced by the Loma Prieta Earthquake. Science 255: 1687-1690.
Risk Management Solutions (RMS), 2006. Managing Tsunami Risk in the Aftermath of the 2004 Indian Ocean
Earthquake and Tsunami. http://www.rms.com/publications
Sahin, M. and Tari, E., 2000. The August 17 Kocaeli and the November 12 Duzce earthquakes in Turkey. Earth
Planets Space 52: 753-757.
Stein, R. S., Barka, A. A., and Dieterich, J. H., 1997. Progressive failure on the North Anatolian fault since
1939 by earthquake stress triggering. Geophysical Journal International 128: 594-604.
Stein, R. S., 2003. Earthquake conversations. Scientific American 288: 72-79.
Swiss Re, 2000. Natural catastrophes and man-made disasters in 1999: Storms and earthquakes lead to the
second-highest losses in insurance history. Sigma No. 2/2000.
Swiss Re, 2004. Natural catastrophes and man-made disasters in 2004: more than 300000 fatalities, record
insured losses. Sigma No 1/2005.
United States Geological Survey (USGS), 2008. Most Destructive Known Earthquakes on Record in the World.
http://earthquake.usgs.gov/regional/world/most_destructive.php
Wesnousky, S. G. and Leffler, L. M., 1992. The repeat time of the 1811 and 1812 New Madrid earthquakes: A
geological perspective. Bulletin of the Seismological Society of America 82: 1,756-1,785.

2008 Risk Management Solutions, Inc.

21